xref: /linux/ipc/mqueue.c (revision f3a8b6645dc2e60d11f20c1c23afd964ff4e55ae)
1 /*
2  * POSIX message queues filesystem for Linux.
3  *
4  * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
5  *                          Michal Wronski          (michal.wronski@gmail.com)
6  *
7  * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
8  * Lockless receive & send, fd based notify:
9  *			    Manfred Spraul	    (manfred@colorfullife.com)
10  *
11  * Audit:                   George Wilson           (ltcgcw@us.ibm.com)
12  *
13  * This file is released under the GPL.
14  */
15 
16 #include <linux/capability.h>
17 #include <linux/init.h>
18 #include <linux/pagemap.h>
19 #include <linux/file.h>
20 #include <linux/mount.h>
21 #include <linux/namei.h>
22 #include <linux/sysctl.h>
23 #include <linux/poll.h>
24 #include <linux/mqueue.h>
25 #include <linux/msg.h>
26 #include <linux/skbuff.h>
27 #include <linux/vmalloc.h>
28 #include <linux/netlink.h>
29 #include <linux/syscalls.h>
30 #include <linux/audit.h>
31 #include <linux/signal.h>
32 #include <linux/mutex.h>
33 #include <linux/nsproxy.h>
34 #include <linux/pid.h>
35 #include <linux/ipc_namespace.h>
36 #include <linux/user_namespace.h>
37 #include <linux/slab.h>
38 
39 #include <net/sock.h>
40 #include "util.h"
41 
42 #define MQUEUE_MAGIC	0x19800202
43 #define DIRENT_SIZE	20
44 #define FILENT_SIZE	80
45 
46 #define SEND		0
47 #define RECV		1
48 
49 #define STATE_NONE	0
50 #define STATE_READY	1
51 
52 struct posix_msg_tree_node {
53 	struct rb_node		rb_node;
54 	struct list_head	msg_list;
55 	int			priority;
56 };
57 
58 struct ext_wait_queue {		/* queue of sleeping tasks */
59 	struct task_struct *task;
60 	struct list_head list;
61 	struct msg_msg *msg;	/* ptr of loaded message */
62 	int state;		/* one of STATE_* values */
63 };
64 
65 struct mqueue_inode_info {
66 	spinlock_t lock;
67 	struct inode vfs_inode;
68 	wait_queue_head_t wait_q;
69 
70 	struct rb_root msg_tree;
71 	struct posix_msg_tree_node *node_cache;
72 	struct mq_attr attr;
73 
74 	struct sigevent notify;
75 	struct pid *notify_owner;
76 	struct user_namespace *notify_user_ns;
77 	struct user_struct *user;	/* user who created, for accounting */
78 	struct sock *notify_sock;
79 	struct sk_buff *notify_cookie;
80 
81 	/* for tasks waiting for free space and messages, respectively */
82 	struct ext_wait_queue e_wait_q[2];
83 
84 	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
85 };
86 
87 static const struct inode_operations mqueue_dir_inode_operations;
88 static const struct file_operations mqueue_file_operations;
89 static const struct super_operations mqueue_super_ops;
90 static void remove_notification(struct mqueue_inode_info *info);
91 
92 static struct kmem_cache *mqueue_inode_cachep;
93 
94 static struct ctl_table_header *mq_sysctl_table;
95 
96 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
97 {
98 	return container_of(inode, struct mqueue_inode_info, vfs_inode);
99 }
100 
101 /*
102  * This routine should be called with the mq_lock held.
103  */
104 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
105 {
106 	return get_ipc_ns(inode->i_sb->s_fs_info);
107 }
108 
109 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
110 {
111 	struct ipc_namespace *ns;
112 
113 	spin_lock(&mq_lock);
114 	ns = __get_ns_from_inode(inode);
115 	spin_unlock(&mq_lock);
116 	return ns;
117 }
118 
119 /* Auxiliary functions to manipulate messages' list */
120 static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
121 {
122 	struct rb_node **p, *parent = NULL;
123 	struct posix_msg_tree_node *leaf;
124 
125 	p = &info->msg_tree.rb_node;
126 	while (*p) {
127 		parent = *p;
128 		leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
129 
130 		if (likely(leaf->priority == msg->m_type))
131 			goto insert_msg;
132 		else if (msg->m_type < leaf->priority)
133 			p = &(*p)->rb_left;
134 		else
135 			p = &(*p)->rb_right;
136 	}
137 	if (info->node_cache) {
138 		leaf = info->node_cache;
139 		info->node_cache = NULL;
140 	} else {
141 		leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
142 		if (!leaf)
143 			return -ENOMEM;
144 		INIT_LIST_HEAD(&leaf->msg_list);
145 	}
146 	leaf->priority = msg->m_type;
147 	rb_link_node(&leaf->rb_node, parent, p);
148 	rb_insert_color(&leaf->rb_node, &info->msg_tree);
149 insert_msg:
150 	info->attr.mq_curmsgs++;
151 	info->qsize += msg->m_ts;
152 	list_add_tail(&msg->m_list, &leaf->msg_list);
153 	return 0;
154 }
155 
156 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
157 {
158 	struct rb_node **p, *parent = NULL;
159 	struct posix_msg_tree_node *leaf;
160 	struct msg_msg *msg;
161 
162 try_again:
163 	p = &info->msg_tree.rb_node;
164 	while (*p) {
165 		parent = *p;
166 		/*
167 		 * During insert, low priorities go to the left and high to the
168 		 * right.  On receive, we want the highest priorities first, so
169 		 * walk all the way to the right.
170 		 */
171 		p = &(*p)->rb_right;
172 	}
173 	if (!parent) {
174 		if (info->attr.mq_curmsgs) {
175 			pr_warn_once("Inconsistency in POSIX message queue, "
176 				     "no tree element, but supposedly messages "
177 				     "should exist!\n");
178 			info->attr.mq_curmsgs = 0;
179 		}
180 		return NULL;
181 	}
182 	leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
183 	if (unlikely(list_empty(&leaf->msg_list))) {
184 		pr_warn_once("Inconsistency in POSIX message queue, "
185 			     "empty leaf node but we haven't implemented "
186 			     "lazy leaf delete!\n");
187 		rb_erase(&leaf->rb_node, &info->msg_tree);
188 		if (info->node_cache) {
189 			kfree(leaf);
190 		} else {
191 			info->node_cache = leaf;
192 		}
193 		goto try_again;
194 	} else {
195 		msg = list_first_entry(&leaf->msg_list,
196 				       struct msg_msg, m_list);
197 		list_del(&msg->m_list);
198 		if (list_empty(&leaf->msg_list)) {
199 			rb_erase(&leaf->rb_node, &info->msg_tree);
200 			if (info->node_cache) {
201 				kfree(leaf);
202 			} else {
203 				info->node_cache = leaf;
204 			}
205 		}
206 	}
207 	info->attr.mq_curmsgs--;
208 	info->qsize -= msg->m_ts;
209 	return msg;
210 }
211 
212 static struct inode *mqueue_get_inode(struct super_block *sb,
213 		struct ipc_namespace *ipc_ns, umode_t mode,
214 		struct mq_attr *attr)
215 {
216 	struct user_struct *u = current_user();
217 	struct inode *inode;
218 	int ret = -ENOMEM;
219 
220 	inode = new_inode(sb);
221 	if (!inode)
222 		goto err;
223 
224 	inode->i_ino = get_next_ino();
225 	inode->i_mode = mode;
226 	inode->i_uid = current_fsuid();
227 	inode->i_gid = current_fsgid();
228 	inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
229 
230 	if (S_ISREG(mode)) {
231 		struct mqueue_inode_info *info;
232 		unsigned long mq_bytes, mq_treesize;
233 
234 		inode->i_fop = &mqueue_file_operations;
235 		inode->i_size = FILENT_SIZE;
236 		/* mqueue specific info */
237 		info = MQUEUE_I(inode);
238 		spin_lock_init(&info->lock);
239 		init_waitqueue_head(&info->wait_q);
240 		INIT_LIST_HEAD(&info->e_wait_q[0].list);
241 		INIT_LIST_HEAD(&info->e_wait_q[1].list);
242 		info->notify_owner = NULL;
243 		info->notify_user_ns = NULL;
244 		info->qsize = 0;
245 		info->user = NULL;	/* set when all is ok */
246 		info->msg_tree = RB_ROOT;
247 		info->node_cache = NULL;
248 		memset(&info->attr, 0, sizeof(info->attr));
249 		info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
250 					   ipc_ns->mq_msg_default);
251 		info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
252 					    ipc_ns->mq_msgsize_default);
253 		if (attr) {
254 			info->attr.mq_maxmsg = attr->mq_maxmsg;
255 			info->attr.mq_msgsize = attr->mq_msgsize;
256 		}
257 		/*
258 		 * We used to allocate a static array of pointers and account
259 		 * the size of that array as well as one msg_msg struct per
260 		 * possible message into the queue size. That's no longer
261 		 * accurate as the queue is now an rbtree and will grow and
262 		 * shrink depending on usage patterns.  We can, however, still
263 		 * account one msg_msg struct per message, but the nodes are
264 		 * allocated depending on priority usage, and most programs
265 		 * only use one, or a handful, of priorities.  However, since
266 		 * this is pinned memory, we need to assume worst case, so
267 		 * that means the min(mq_maxmsg, max_priorities) * struct
268 		 * posix_msg_tree_node.
269 		 */
270 		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
271 			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
272 			sizeof(struct posix_msg_tree_node);
273 
274 		mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
275 					  info->attr.mq_msgsize);
276 
277 		spin_lock(&mq_lock);
278 		if (u->mq_bytes + mq_bytes < u->mq_bytes ||
279 		    u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
280 			spin_unlock(&mq_lock);
281 			/* mqueue_evict_inode() releases info->messages */
282 			ret = -EMFILE;
283 			goto out_inode;
284 		}
285 		u->mq_bytes += mq_bytes;
286 		spin_unlock(&mq_lock);
287 
288 		/* all is ok */
289 		info->user = get_uid(u);
290 	} else if (S_ISDIR(mode)) {
291 		inc_nlink(inode);
292 		/* Some things misbehave if size == 0 on a directory */
293 		inode->i_size = 2 * DIRENT_SIZE;
294 		inode->i_op = &mqueue_dir_inode_operations;
295 		inode->i_fop = &simple_dir_operations;
296 	}
297 
298 	return inode;
299 out_inode:
300 	iput(inode);
301 err:
302 	return ERR_PTR(ret);
303 }
304 
305 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
306 {
307 	struct inode *inode;
308 	struct ipc_namespace *ns = sb->s_fs_info;
309 
310 	sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
311 	sb->s_blocksize = PAGE_SIZE;
312 	sb->s_blocksize_bits = PAGE_SHIFT;
313 	sb->s_magic = MQUEUE_MAGIC;
314 	sb->s_op = &mqueue_super_ops;
315 
316 	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
317 	if (IS_ERR(inode))
318 		return PTR_ERR(inode);
319 
320 	sb->s_root = d_make_root(inode);
321 	if (!sb->s_root)
322 		return -ENOMEM;
323 	return 0;
324 }
325 
326 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
327 			 int flags, const char *dev_name,
328 			 void *data)
329 {
330 	struct ipc_namespace *ns;
331 	if (flags & MS_KERNMOUNT) {
332 		ns = data;
333 		data = NULL;
334 	} else {
335 		ns = current->nsproxy->ipc_ns;
336 	}
337 	return mount_ns(fs_type, flags, data, ns, ns->user_ns, mqueue_fill_super);
338 }
339 
340 static void init_once(void *foo)
341 {
342 	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
343 
344 	inode_init_once(&p->vfs_inode);
345 }
346 
347 static struct inode *mqueue_alloc_inode(struct super_block *sb)
348 {
349 	struct mqueue_inode_info *ei;
350 
351 	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
352 	if (!ei)
353 		return NULL;
354 	return &ei->vfs_inode;
355 }
356 
357 static void mqueue_i_callback(struct rcu_head *head)
358 {
359 	struct inode *inode = container_of(head, struct inode, i_rcu);
360 	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
361 }
362 
363 static void mqueue_destroy_inode(struct inode *inode)
364 {
365 	call_rcu(&inode->i_rcu, mqueue_i_callback);
366 }
367 
368 static void mqueue_evict_inode(struct inode *inode)
369 {
370 	struct mqueue_inode_info *info;
371 	struct user_struct *user;
372 	unsigned long mq_bytes, mq_treesize;
373 	struct ipc_namespace *ipc_ns;
374 	struct msg_msg *msg;
375 
376 	clear_inode(inode);
377 
378 	if (S_ISDIR(inode->i_mode))
379 		return;
380 
381 	ipc_ns = get_ns_from_inode(inode);
382 	info = MQUEUE_I(inode);
383 	spin_lock(&info->lock);
384 	while ((msg = msg_get(info)) != NULL)
385 		free_msg(msg);
386 	kfree(info->node_cache);
387 	spin_unlock(&info->lock);
388 
389 	/* Total amount of bytes accounted for the mqueue */
390 	mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
391 		min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
392 		sizeof(struct posix_msg_tree_node);
393 
394 	mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
395 				  info->attr.mq_msgsize);
396 
397 	user = info->user;
398 	if (user) {
399 		spin_lock(&mq_lock);
400 		user->mq_bytes -= mq_bytes;
401 		/*
402 		 * get_ns_from_inode() ensures that the
403 		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
404 		 * to which we now hold a reference, or it is NULL.
405 		 * We can't put it here under mq_lock, though.
406 		 */
407 		if (ipc_ns)
408 			ipc_ns->mq_queues_count--;
409 		spin_unlock(&mq_lock);
410 		free_uid(user);
411 	}
412 	if (ipc_ns)
413 		put_ipc_ns(ipc_ns);
414 }
415 
416 static int mqueue_create(struct inode *dir, struct dentry *dentry,
417 				umode_t mode, bool excl)
418 {
419 	struct inode *inode;
420 	struct mq_attr *attr = dentry->d_fsdata;
421 	int error;
422 	struct ipc_namespace *ipc_ns;
423 
424 	spin_lock(&mq_lock);
425 	ipc_ns = __get_ns_from_inode(dir);
426 	if (!ipc_ns) {
427 		error = -EACCES;
428 		goto out_unlock;
429 	}
430 
431 	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
432 	    !capable(CAP_SYS_RESOURCE)) {
433 		error = -ENOSPC;
434 		goto out_unlock;
435 	}
436 	ipc_ns->mq_queues_count++;
437 	spin_unlock(&mq_lock);
438 
439 	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
440 	if (IS_ERR(inode)) {
441 		error = PTR_ERR(inode);
442 		spin_lock(&mq_lock);
443 		ipc_ns->mq_queues_count--;
444 		goto out_unlock;
445 	}
446 
447 	put_ipc_ns(ipc_ns);
448 	dir->i_size += DIRENT_SIZE;
449 	dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
450 
451 	d_instantiate(dentry, inode);
452 	dget(dentry);
453 	return 0;
454 out_unlock:
455 	spin_unlock(&mq_lock);
456 	if (ipc_ns)
457 		put_ipc_ns(ipc_ns);
458 	return error;
459 }
460 
461 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
462 {
463 	struct inode *inode = d_inode(dentry);
464 
465 	dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
466 	dir->i_size -= DIRENT_SIZE;
467 	drop_nlink(inode);
468 	dput(dentry);
469 	return 0;
470 }
471 
472 /*
473 *	This is routine for system read from queue file.
474 *	To avoid mess with doing here some sort of mq_receive we allow
475 *	to read only queue size & notification info (the only values
476 *	that are interesting from user point of view and aren't accessible
477 *	through std routines)
478 */
479 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
480 				size_t count, loff_t *off)
481 {
482 	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
483 	char buffer[FILENT_SIZE];
484 	ssize_t ret;
485 
486 	spin_lock(&info->lock);
487 	snprintf(buffer, sizeof(buffer),
488 			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
489 			info->qsize,
490 			info->notify_owner ? info->notify.sigev_notify : 0,
491 			(info->notify_owner &&
492 			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
493 				info->notify.sigev_signo : 0,
494 			pid_vnr(info->notify_owner));
495 	spin_unlock(&info->lock);
496 	buffer[sizeof(buffer)-1] = '\0';
497 
498 	ret = simple_read_from_buffer(u_data, count, off, buffer,
499 				strlen(buffer));
500 	if (ret <= 0)
501 		return ret;
502 
503 	file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
504 	return ret;
505 }
506 
507 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
508 {
509 	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
510 
511 	spin_lock(&info->lock);
512 	if (task_tgid(current) == info->notify_owner)
513 		remove_notification(info);
514 
515 	spin_unlock(&info->lock);
516 	return 0;
517 }
518 
519 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
520 {
521 	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
522 	int retval = 0;
523 
524 	poll_wait(filp, &info->wait_q, poll_tab);
525 
526 	spin_lock(&info->lock);
527 	if (info->attr.mq_curmsgs)
528 		retval = POLLIN | POLLRDNORM;
529 
530 	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
531 		retval |= POLLOUT | POLLWRNORM;
532 	spin_unlock(&info->lock);
533 
534 	return retval;
535 }
536 
537 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
538 static void wq_add(struct mqueue_inode_info *info, int sr,
539 			struct ext_wait_queue *ewp)
540 {
541 	struct ext_wait_queue *walk;
542 
543 	ewp->task = current;
544 
545 	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
546 		if (walk->task->static_prio <= current->static_prio) {
547 			list_add_tail(&ewp->list, &walk->list);
548 			return;
549 		}
550 	}
551 	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
552 }
553 
554 /*
555  * Puts current task to sleep. Caller must hold queue lock. After return
556  * lock isn't held.
557  * sr: SEND or RECV
558  */
559 static int wq_sleep(struct mqueue_inode_info *info, int sr,
560 		    ktime_t *timeout, struct ext_wait_queue *ewp)
561 {
562 	int retval;
563 	signed long time;
564 
565 	wq_add(info, sr, ewp);
566 
567 	for (;;) {
568 		__set_current_state(TASK_INTERRUPTIBLE);
569 
570 		spin_unlock(&info->lock);
571 		time = schedule_hrtimeout_range_clock(timeout, 0,
572 			HRTIMER_MODE_ABS, CLOCK_REALTIME);
573 
574 		if (ewp->state == STATE_READY) {
575 			retval = 0;
576 			goto out;
577 		}
578 		spin_lock(&info->lock);
579 		if (ewp->state == STATE_READY) {
580 			retval = 0;
581 			goto out_unlock;
582 		}
583 		if (signal_pending(current)) {
584 			retval = -ERESTARTSYS;
585 			break;
586 		}
587 		if (time == 0) {
588 			retval = -ETIMEDOUT;
589 			break;
590 		}
591 	}
592 	list_del(&ewp->list);
593 out_unlock:
594 	spin_unlock(&info->lock);
595 out:
596 	return retval;
597 }
598 
599 /*
600  * Returns waiting task that should be serviced first or NULL if none exists
601  */
602 static struct ext_wait_queue *wq_get_first_waiter(
603 		struct mqueue_inode_info *info, int sr)
604 {
605 	struct list_head *ptr;
606 
607 	ptr = info->e_wait_q[sr].list.prev;
608 	if (ptr == &info->e_wait_q[sr].list)
609 		return NULL;
610 	return list_entry(ptr, struct ext_wait_queue, list);
611 }
612 
613 
614 static inline void set_cookie(struct sk_buff *skb, char code)
615 {
616 	((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
617 }
618 
619 /*
620  * The next function is only to split too long sys_mq_timedsend
621  */
622 static void __do_notify(struct mqueue_inode_info *info)
623 {
624 	/* notification
625 	 * invoked when there is registered process and there isn't process
626 	 * waiting synchronously for message AND state of queue changed from
627 	 * empty to not empty. Here we are sure that no one is waiting
628 	 * synchronously. */
629 	if (info->notify_owner &&
630 	    info->attr.mq_curmsgs == 1) {
631 		struct siginfo sig_i;
632 		switch (info->notify.sigev_notify) {
633 		case SIGEV_NONE:
634 			break;
635 		case SIGEV_SIGNAL:
636 			/* sends signal */
637 
638 			sig_i.si_signo = info->notify.sigev_signo;
639 			sig_i.si_errno = 0;
640 			sig_i.si_code = SI_MESGQ;
641 			sig_i.si_value = info->notify.sigev_value;
642 			/* map current pid/uid into info->owner's namespaces */
643 			rcu_read_lock();
644 			sig_i.si_pid = task_tgid_nr_ns(current,
645 						ns_of_pid(info->notify_owner));
646 			sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
647 			rcu_read_unlock();
648 
649 			kill_pid_info(info->notify.sigev_signo,
650 				      &sig_i, info->notify_owner);
651 			break;
652 		case SIGEV_THREAD:
653 			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
654 			netlink_sendskb(info->notify_sock, info->notify_cookie);
655 			break;
656 		}
657 		/* after notification unregisters process */
658 		put_pid(info->notify_owner);
659 		put_user_ns(info->notify_user_ns);
660 		info->notify_owner = NULL;
661 		info->notify_user_ns = NULL;
662 	}
663 	wake_up(&info->wait_q);
664 }
665 
666 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
667 			   ktime_t *expires, struct timespec *ts)
668 {
669 	if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
670 		return -EFAULT;
671 	if (!timespec_valid(ts))
672 		return -EINVAL;
673 
674 	*expires = timespec_to_ktime(*ts);
675 	return 0;
676 }
677 
678 static void remove_notification(struct mqueue_inode_info *info)
679 {
680 	if (info->notify_owner != NULL &&
681 	    info->notify.sigev_notify == SIGEV_THREAD) {
682 		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
683 		netlink_sendskb(info->notify_sock, info->notify_cookie);
684 	}
685 	put_pid(info->notify_owner);
686 	put_user_ns(info->notify_user_ns);
687 	info->notify_owner = NULL;
688 	info->notify_user_ns = NULL;
689 }
690 
691 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
692 {
693 	int mq_treesize;
694 	unsigned long total_size;
695 
696 	if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
697 		return -EINVAL;
698 	if (capable(CAP_SYS_RESOURCE)) {
699 		if (attr->mq_maxmsg > HARD_MSGMAX ||
700 		    attr->mq_msgsize > HARD_MSGSIZEMAX)
701 			return -EINVAL;
702 	} else {
703 		if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
704 				attr->mq_msgsize > ipc_ns->mq_msgsize_max)
705 			return -EINVAL;
706 	}
707 	/* check for overflow */
708 	if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
709 		return -EOVERFLOW;
710 	mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
711 		min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
712 		sizeof(struct posix_msg_tree_node);
713 	total_size = attr->mq_maxmsg * attr->mq_msgsize;
714 	if (total_size + mq_treesize < total_size)
715 		return -EOVERFLOW;
716 	return 0;
717 }
718 
719 /*
720  * Invoked when creating a new queue via sys_mq_open
721  */
722 static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
723 			struct path *path, int oflag, umode_t mode,
724 			struct mq_attr *attr)
725 {
726 	const struct cred *cred = current_cred();
727 	int ret;
728 
729 	if (attr) {
730 		ret = mq_attr_ok(ipc_ns, attr);
731 		if (ret)
732 			return ERR_PTR(ret);
733 		/* store for use during create */
734 		path->dentry->d_fsdata = attr;
735 	} else {
736 		struct mq_attr def_attr;
737 
738 		def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
739 					 ipc_ns->mq_msg_default);
740 		def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
741 					  ipc_ns->mq_msgsize_default);
742 		ret = mq_attr_ok(ipc_ns, &def_attr);
743 		if (ret)
744 			return ERR_PTR(ret);
745 	}
746 
747 	mode &= ~current_umask();
748 	ret = vfs_create(dir, path->dentry, mode, true);
749 	path->dentry->d_fsdata = NULL;
750 	if (ret)
751 		return ERR_PTR(ret);
752 	return dentry_open(path, oflag, cred);
753 }
754 
755 /* Opens existing queue */
756 static struct file *do_open(struct path *path, int oflag)
757 {
758 	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
759 						  MAY_READ | MAY_WRITE };
760 	int acc;
761 	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
762 		return ERR_PTR(-EINVAL);
763 	acc = oflag2acc[oflag & O_ACCMODE];
764 	if (inode_permission(d_inode(path->dentry), acc))
765 		return ERR_PTR(-EACCES);
766 	return dentry_open(path, oflag, current_cred());
767 }
768 
769 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
770 		struct mq_attr __user *, u_attr)
771 {
772 	struct path path;
773 	struct file *filp;
774 	struct filename *name;
775 	struct mq_attr attr;
776 	int fd, error;
777 	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
778 	struct vfsmount *mnt = ipc_ns->mq_mnt;
779 	struct dentry *root = mnt->mnt_root;
780 	int ro;
781 
782 	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
783 		return -EFAULT;
784 
785 	audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
786 
787 	if (IS_ERR(name = getname(u_name)))
788 		return PTR_ERR(name);
789 
790 	fd = get_unused_fd_flags(O_CLOEXEC);
791 	if (fd < 0)
792 		goto out_putname;
793 
794 	ro = mnt_want_write(mnt);	/* we'll drop it in any case */
795 	error = 0;
796 	inode_lock(d_inode(root));
797 	path.dentry = lookup_one_len(name->name, root, strlen(name->name));
798 	if (IS_ERR(path.dentry)) {
799 		error = PTR_ERR(path.dentry);
800 		goto out_putfd;
801 	}
802 	path.mnt = mntget(mnt);
803 
804 	if (oflag & O_CREAT) {
805 		if (d_really_is_positive(path.dentry)) {	/* entry already exists */
806 			audit_inode(name, path.dentry, 0);
807 			if (oflag & O_EXCL) {
808 				error = -EEXIST;
809 				goto out;
810 			}
811 			filp = do_open(&path, oflag);
812 		} else {
813 			if (ro) {
814 				error = ro;
815 				goto out;
816 			}
817 			audit_inode_parent_hidden(name, root);
818 			filp = do_create(ipc_ns, d_inode(root),
819 						&path, oflag, mode,
820 						u_attr ? &attr : NULL);
821 		}
822 	} else {
823 		if (d_really_is_negative(path.dentry)) {
824 			error = -ENOENT;
825 			goto out;
826 		}
827 		audit_inode(name, path.dentry, 0);
828 		filp = do_open(&path, oflag);
829 	}
830 
831 	if (!IS_ERR(filp))
832 		fd_install(fd, filp);
833 	else
834 		error = PTR_ERR(filp);
835 out:
836 	path_put(&path);
837 out_putfd:
838 	if (error) {
839 		put_unused_fd(fd);
840 		fd = error;
841 	}
842 	inode_unlock(d_inode(root));
843 	if (!ro)
844 		mnt_drop_write(mnt);
845 out_putname:
846 	putname(name);
847 	return fd;
848 }
849 
850 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
851 {
852 	int err;
853 	struct filename *name;
854 	struct dentry *dentry;
855 	struct inode *inode = NULL;
856 	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
857 	struct vfsmount *mnt = ipc_ns->mq_mnt;
858 
859 	name = getname(u_name);
860 	if (IS_ERR(name))
861 		return PTR_ERR(name);
862 
863 	audit_inode_parent_hidden(name, mnt->mnt_root);
864 	err = mnt_want_write(mnt);
865 	if (err)
866 		goto out_name;
867 	inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
868 	dentry = lookup_one_len(name->name, mnt->mnt_root,
869 				strlen(name->name));
870 	if (IS_ERR(dentry)) {
871 		err = PTR_ERR(dentry);
872 		goto out_unlock;
873 	}
874 
875 	inode = d_inode(dentry);
876 	if (!inode) {
877 		err = -ENOENT;
878 	} else {
879 		ihold(inode);
880 		err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
881 	}
882 	dput(dentry);
883 
884 out_unlock:
885 	inode_unlock(d_inode(mnt->mnt_root));
886 	if (inode)
887 		iput(inode);
888 	mnt_drop_write(mnt);
889 out_name:
890 	putname(name);
891 
892 	return err;
893 }
894 
895 /* Pipelined send and receive functions.
896  *
897  * If a receiver finds no waiting message, then it registers itself in the
898  * list of waiting receivers. A sender checks that list before adding the new
899  * message into the message array. If there is a waiting receiver, then it
900  * bypasses the message array and directly hands the message over to the
901  * receiver. The receiver accepts the message and returns without grabbing the
902  * queue spinlock:
903  *
904  * - Set pointer to message.
905  * - Queue the receiver task for later wakeup (without the info->lock).
906  * - Update its state to STATE_READY. Now the receiver can continue.
907  * - Wake up the process after the lock is dropped. Should the process wake up
908  *   before this wakeup (due to a timeout or a signal) it will either see
909  *   STATE_READY and continue or acquire the lock to check the state again.
910  *
911  * The same algorithm is used for senders.
912  */
913 
914 /* pipelined_send() - send a message directly to the task waiting in
915  * sys_mq_timedreceive() (without inserting message into a queue).
916  */
917 static inline void pipelined_send(struct wake_q_head *wake_q,
918 				  struct mqueue_inode_info *info,
919 				  struct msg_msg *message,
920 				  struct ext_wait_queue *receiver)
921 {
922 	receiver->msg = message;
923 	list_del(&receiver->list);
924 	wake_q_add(wake_q, receiver->task);
925 	/*
926 	 * Rely on the implicit cmpxchg barrier from wake_q_add such
927 	 * that we can ensure that updating receiver->state is the last
928 	 * write operation: As once set, the receiver can continue,
929 	 * and if we don't have the reference count from the wake_q,
930 	 * yet, at that point we can later have a use-after-free
931 	 * condition and bogus wakeup.
932 	 */
933 	receiver->state = STATE_READY;
934 }
935 
936 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
937  * gets its message and put to the queue (we have one free place for sure). */
938 static inline void pipelined_receive(struct wake_q_head *wake_q,
939 				     struct mqueue_inode_info *info)
940 {
941 	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
942 
943 	if (!sender) {
944 		/* for poll */
945 		wake_up_interruptible(&info->wait_q);
946 		return;
947 	}
948 	if (msg_insert(sender->msg, info))
949 		return;
950 
951 	list_del(&sender->list);
952 	wake_q_add(wake_q, sender->task);
953 	sender->state = STATE_READY;
954 }
955 
956 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
957 		size_t, msg_len, unsigned int, msg_prio,
958 		const struct timespec __user *, u_abs_timeout)
959 {
960 	struct fd f;
961 	struct inode *inode;
962 	struct ext_wait_queue wait;
963 	struct ext_wait_queue *receiver;
964 	struct msg_msg *msg_ptr;
965 	struct mqueue_inode_info *info;
966 	ktime_t expires, *timeout = NULL;
967 	struct timespec ts;
968 	struct posix_msg_tree_node *new_leaf = NULL;
969 	int ret = 0;
970 	WAKE_Q(wake_q);
971 
972 	if (u_abs_timeout) {
973 		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
974 		if (res)
975 			return res;
976 		timeout = &expires;
977 	}
978 
979 	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
980 		return -EINVAL;
981 
982 	audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
983 
984 	f = fdget(mqdes);
985 	if (unlikely(!f.file)) {
986 		ret = -EBADF;
987 		goto out;
988 	}
989 
990 	inode = file_inode(f.file);
991 	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
992 		ret = -EBADF;
993 		goto out_fput;
994 	}
995 	info = MQUEUE_I(inode);
996 	audit_file(f.file);
997 
998 	if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
999 		ret = -EBADF;
1000 		goto out_fput;
1001 	}
1002 
1003 	if (unlikely(msg_len > info->attr.mq_msgsize)) {
1004 		ret = -EMSGSIZE;
1005 		goto out_fput;
1006 	}
1007 
1008 	/* First try to allocate memory, before doing anything with
1009 	 * existing queues. */
1010 	msg_ptr = load_msg(u_msg_ptr, msg_len);
1011 	if (IS_ERR(msg_ptr)) {
1012 		ret = PTR_ERR(msg_ptr);
1013 		goto out_fput;
1014 	}
1015 	msg_ptr->m_ts = msg_len;
1016 	msg_ptr->m_type = msg_prio;
1017 
1018 	/*
1019 	 * msg_insert really wants us to have a valid, spare node struct so
1020 	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1021 	 * fall back to that if necessary.
1022 	 */
1023 	if (!info->node_cache)
1024 		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1025 
1026 	spin_lock(&info->lock);
1027 
1028 	if (!info->node_cache && new_leaf) {
1029 		/* Save our speculative allocation into the cache */
1030 		INIT_LIST_HEAD(&new_leaf->msg_list);
1031 		info->node_cache = new_leaf;
1032 		new_leaf = NULL;
1033 	} else {
1034 		kfree(new_leaf);
1035 	}
1036 
1037 	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1038 		if (f.file->f_flags & O_NONBLOCK) {
1039 			ret = -EAGAIN;
1040 		} else {
1041 			wait.task = current;
1042 			wait.msg = (void *) msg_ptr;
1043 			wait.state = STATE_NONE;
1044 			ret = wq_sleep(info, SEND, timeout, &wait);
1045 			/*
1046 			 * wq_sleep must be called with info->lock held, and
1047 			 * returns with the lock released
1048 			 */
1049 			goto out_free;
1050 		}
1051 	} else {
1052 		receiver = wq_get_first_waiter(info, RECV);
1053 		if (receiver) {
1054 			pipelined_send(&wake_q, info, msg_ptr, receiver);
1055 		} else {
1056 			/* adds message to the queue */
1057 			ret = msg_insert(msg_ptr, info);
1058 			if (ret)
1059 				goto out_unlock;
1060 			__do_notify(info);
1061 		}
1062 		inode->i_atime = inode->i_mtime = inode->i_ctime =
1063 				current_time(inode);
1064 	}
1065 out_unlock:
1066 	spin_unlock(&info->lock);
1067 	wake_up_q(&wake_q);
1068 out_free:
1069 	if (ret)
1070 		free_msg(msg_ptr);
1071 out_fput:
1072 	fdput(f);
1073 out:
1074 	return ret;
1075 }
1076 
1077 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1078 		size_t, msg_len, unsigned int __user *, u_msg_prio,
1079 		const struct timespec __user *, u_abs_timeout)
1080 {
1081 	ssize_t ret;
1082 	struct msg_msg *msg_ptr;
1083 	struct fd f;
1084 	struct inode *inode;
1085 	struct mqueue_inode_info *info;
1086 	struct ext_wait_queue wait;
1087 	ktime_t expires, *timeout = NULL;
1088 	struct timespec ts;
1089 	struct posix_msg_tree_node *new_leaf = NULL;
1090 
1091 	if (u_abs_timeout) {
1092 		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1093 		if (res)
1094 			return res;
1095 		timeout = &expires;
1096 	}
1097 
1098 	audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1099 
1100 	f = fdget(mqdes);
1101 	if (unlikely(!f.file)) {
1102 		ret = -EBADF;
1103 		goto out;
1104 	}
1105 
1106 	inode = file_inode(f.file);
1107 	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1108 		ret = -EBADF;
1109 		goto out_fput;
1110 	}
1111 	info = MQUEUE_I(inode);
1112 	audit_file(f.file);
1113 
1114 	if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1115 		ret = -EBADF;
1116 		goto out_fput;
1117 	}
1118 
1119 	/* checks if buffer is big enough */
1120 	if (unlikely(msg_len < info->attr.mq_msgsize)) {
1121 		ret = -EMSGSIZE;
1122 		goto out_fput;
1123 	}
1124 
1125 	/*
1126 	 * msg_insert really wants us to have a valid, spare node struct so
1127 	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1128 	 * fall back to that if necessary.
1129 	 */
1130 	if (!info->node_cache)
1131 		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1132 
1133 	spin_lock(&info->lock);
1134 
1135 	if (!info->node_cache && new_leaf) {
1136 		/* Save our speculative allocation into the cache */
1137 		INIT_LIST_HEAD(&new_leaf->msg_list);
1138 		info->node_cache = new_leaf;
1139 	} else {
1140 		kfree(new_leaf);
1141 	}
1142 
1143 	if (info->attr.mq_curmsgs == 0) {
1144 		if (f.file->f_flags & O_NONBLOCK) {
1145 			spin_unlock(&info->lock);
1146 			ret = -EAGAIN;
1147 		} else {
1148 			wait.task = current;
1149 			wait.state = STATE_NONE;
1150 			ret = wq_sleep(info, RECV, timeout, &wait);
1151 			msg_ptr = wait.msg;
1152 		}
1153 	} else {
1154 		WAKE_Q(wake_q);
1155 
1156 		msg_ptr = msg_get(info);
1157 
1158 		inode->i_atime = inode->i_mtime = inode->i_ctime =
1159 				current_time(inode);
1160 
1161 		/* There is now free space in queue. */
1162 		pipelined_receive(&wake_q, info);
1163 		spin_unlock(&info->lock);
1164 		wake_up_q(&wake_q);
1165 		ret = 0;
1166 	}
1167 	if (ret == 0) {
1168 		ret = msg_ptr->m_ts;
1169 
1170 		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1171 			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1172 			ret = -EFAULT;
1173 		}
1174 		free_msg(msg_ptr);
1175 	}
1176 out_fput:
1177 	fdput(f);
1178 out:
1179 	return ret;
1180 }
1181 
1182 /*
1183  * Notes: the case when user wants us to deregister (with NULL as pointer)
1184  * and he isn't currently owner of notification, will be silently discarded.
1185  * It isn't explicitly defined in the POSIX.
1186  */
1187 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1188 		const struct sigevent __user *, u_notification)
1189 {
1190 	int ret;
1191 	struct fd f;
1192 	struct sock *sock;
1193 	struct inode *inode;
1194 	struct sigevent notification;
1195 	struct mqueue_inode_info *info;
1196 	struct sk_buff *nc;
1197 
1198 	if (u_notification) {
1199 		if (copy_from_user(&notification, u_notification,
1200 					sizeof(struct sigevent)))
1201 			return -EFAULT;
1202 	}
1203 
1204 	audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1205 
1206 	nc = NULL;
1207 	sock = NULL;
1208 	if (u_notification != NULL) {
1209 		if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1210 			     notification.sigev_notify != SIGEV_SIGNAL &&
1211 			     notification.sigev_notify != SIGEV_THREAD))
1212 			return -EINVAL;
1213 		if (notification.sigev_notify == SIGEV_SIGNAL &&
1214 			!valid_signal(notification.sigev_signo)) {
1215 			return -EINVAL;
1216 		}
1217 		if (notification.sigev_notify == SIGEV_THREAD) {
1218 			long timeo;
1219 
1220 			/* create the notify skb */
1221 			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1222 			if (!nc) {
1223 				ret = -ENOMEM;
1224 				goto out;
1225 			}
1226 			if (copy_from_user(nc->data,
1227 					notification.sigev_value.sival_ptr,
1228 					NOTIFY_COOKIE_LEN)) {
1229 				ret = -EFAULT;
1230 				goto out;
1231 			}
1232 
1233 			/* TODO: add a header? */
1234 			skb_put(nc, NOTIFY_COOKIE_LEN);
1235 			/* and attach it to the socket */
1236 retry:
1237 			f = fdget(notification.sigev_signo);
1238 			if (!f.file) {
1239 				ret = -EBADF;
1240 				goto out;
1241 			}
1242 			sock = netlink_getsockbyfilp(f.file);
1243 			fdput(f);
1244 			if (IS_ERR(sock)) {
1245 				ret = PTR_ERR(sock);
1246 				sock = NULL;
1247 				goto out;
1248 			}
1249 
1250 			timeo = MAX_SCHEDULE_TIMEOUT;
1251 			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1252 			if (ret == 1)
1253 				goto retry;
1254 			if (ret) {
1255 				sock = NULL;
1256 				nc = NULL;
1257 				goto out;
1258 			}
1259 		}
1260 	}
1261 
1262 	f = fdget(mqdes);
1263 	if (!f.file) {
1264 		ret = -EBADF;
1265 		goto out;
1266 	}
1267 
1268 	inode = file_inode(f.file);
1269 	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1270 		ret = -EBADF;
1271 		goto out_fput;
1272 	}
1273 	info = MQUEUE_I(inode);
1274 
1275 	ret = 0;
1276 	spin_lock(&info->lock);
1277 	if (u_notification == NULL) {
1278 		if (info->notify_owner == task_tgid(current)) {
1279 			remove_notification(info);
1280 			inode->i_atime = inode->i_ctime = current_time(inode);
1281 		}
1282 	} else if (info->notify_owner != NULL) {
1283 		ret = -EBUSY;
1284 	} else {
1285 		switch (notification.sigev_notify) {
1286 		case SIGEV_NONE:
1287 			info->notify.sigev_notify = SIGEV_NONE;
1288 			break;
1289 		case SIGEV_THREAD:
1290 			info->notify_sock = sock;
1291 			info->notify_cookie = nc;
1292 			sock = NULL;
1293 			nc = NULL;
1294 			info->notify.sigev_notify = SIGEV_THREAD;
1295 			break;
1296 		case SIGEV_SIGNAL:
1297 			info->notify.sigev_signo = notification.sigev_signo;
1298 			info->notify.sigev_value = notification.sigev_value;
1299 			info->notify.sigev_notify = SIGEV_SIGNAL;
1300 			break;
1301 		}
1302 
1303 		info->notify_owner = get_pid(task_tgid(current));
1304 		info->notify_user_ns = get_user_ns(current_user_ns());
1305 		inode->i_atime = inode->i_ctime = current_time(inode);
1306 	}
1307 	spin_unlock(&info->lock);
1308 out_fput:
1309 	fdput(f);
1310 out:
1311 	if (sock)
1312 		netlink_detachskb(sock, nc);
1313 	else if (nc)
1314 		dev_kfree_skb(nc);
1315 
1316 	return ret;
1317 }
1318 
1319 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1320 		const struct mq_attr __user *, u_mqstat,
1321 		struct mq_attr __user *, u_omqstat)
1322 {
1323 	int ret;
1324 	struct mq_attr mqstat, omqstat;
1325 	struct fd f;
1326 	struct inode *inode;
1327 	struct mqueue_inode_info *info;
1328 
1329 	if (u_mqstat != NULL) {
1330 		if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1331 			return -EFAULT;
1332 		if (mqstat.mq_flags & (~O_NONBLOCK))
1333 			return -EINVAL;
1334 	}
1335 
1336 	f = fdget(mqdes);
1337 	if (!f.file) {
1338 		ret = -EBADF;
1339 		goto out;
1340 	}
1341 
1342 	inode = file_inode(f.file);
1343 	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1344 		ret = -EBADF;
1345 		goto out_fput;
1346 	}
1347 	info = MQUEUE_I(inode);
1348 
1349 	spin_lock(&info->lock);
1350 
1351 	omqstat = info->attr;
1352 	omqstat.mq_flags = f.file->f_flags & O_NONBLOCK;
1353 	if (u_mqstat) {
1354 		audit_mq_getsetattr(mqdes, &mqstat);
1355 		spin_lock(&f.file->f_lock);
1356 		if (mqstat.mq_flags & O_NONBLOCK)
1357 			f.file->f_flags |= O_NONBLOCK;
1358 		else
1359 			f.file->f_flags &= ~O_NONBLOCK;
1360 		spin_unlock(&f.file->f_lock);
1361 
1362 		inode->i_atime = inode->i_ctime = current_time(inode);
1363 	}
1364 
1365 	spin_unlock(&info->lock);
1366 
1367 	ret = 0;
1368 	if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1369 						sizeof(struct mq_attr)))
1370 		ret = -EFAULT;
1371 
1372 out_fput:
1373 	fdput(f);
1374 out:
1375 	return ret;
1376 }
1377 
1378 static const struct inode_operations mqueue_dir_inode_operations = {
1379 	.lookup = simple_lookup,
1380 	.create = mqueue_create,
1381 	.unlink = mqueue_unlink,
1382 };
1383 
1384 static const struct file_operations mqueue_file_operations = {
1385 	.flush = mqueue_flush_file,
1386 	.poll = mqueue_poll_file,
1387 	.read = mqueue_read_file,
1388 	.llseek = default_llseek,
1389 };
1390 
1391 static const struct super_operations mqueue_super_ops = {
1392 	.alloc_inode = mqueue_alloc_inode,
1393 	.destroy_inode = mqueue_destroy_inode,
1394 	.evict_inode = mqueue_evict_inode,
1395 	.statfs = simple_statfs,
1396 };
1397 
1398 static struct file_system_type mqueue_fs_type = {
1399 	.name = "mqueue",
1400 	.mount = mqueue_mount,
1401 	.kill_sb = kill_litter_super,
1402 	.fs_flags = FS_USERNS_MOUNT,
1403 };
1404 
1405 int mq_init_ns(struct ipc_namespace *ns)
1406 {
1407 	ns->mq_queues_count  = 0;
1408 	ns->mq_queues_max    = DFLT_QUEUESMAX;
1409 	ns->mq_msg_max       = DFLT_MSGMAX;
1410 	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1411 	ns->mq_msg_default   = DFLT_MSG;
1412 	ns->mq_msgsize_default  = DFLT_MSGSIZE;
1413 
1414 	ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1415 	if (IS_ERR(ns->mq_mnt)) {
1416 		int err = PTR_ERR(ns->mq_mnt);
1417 		ns->mq_mnt = NULL;
1418 		return err;
1419 	}
1420 	return 0;
1421 }
1422 
1423 void mq_clear_sbinfo(struct ipc_namespace *ns)
1424 {
1425 	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1426 }
1427 
1428 void mq_put_mnt(struct ipc_namespace *ns)
1429 {
1430 	kern_unmount(ns->mq_mnt);
1431 }
1432 
1433 static int __init init_mqueue_fs(void)
1434 {
1435 	int error;
1436 
1437 	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1438 				sizeof(struct mqueue_inode_info), 0,
1439 				SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1440 	if (mqueue_inode_cachep == NULL)
1441 		return -ENOMEM;
1442 
1443 	/* ignore failures - they are not fatal */
1444 	mq_sysctl_table = mq_register_sysctl_table();
1445 
1446 	error = register_filesystem(&mqueue_fs_type);
1447 	if (error)
1448 		goto out_sysctl;
1449 
1450 	spin_lock_init(&mq_lock);
1451 
1452 	error = mq_init_ns(&init_ipc_ns);
1453 	if (error)
1454 		goto out_filesystem;
1455 
1456 	return 0;
1457 
1458 out_filesystem:
1459 	unregister_filesystem(&mqueue_fs_type);
1460 out_sysctl:
1461 	if (mq_sysctl_table)
1462 		unregister_sysctl_table(mq_sysctl_table);
1463 	kmem_cache_destroy(mqueue_inode_cachep);
1464 	return error;
1465 }
1466 
1467 device_initcall(init_mqueue_fs);
1468